Charted: Weather Still Takes 61% of US ATC Delay Minutes — Staffing Is Only 4%, But Seven Facilities Own 86% of It

The gate agent blames staffing. Your seatmate blames the weather. The federal stopwatch — kept since long before either of you reached the airport — disagrees with both of them, quietly and in writing. When flights stack on the taxiway or hold for an EDCT, the FAA's Operations Network (OPSNET) records a reportable delay: instrument-flight-rules traffic detained at least 15 minutes by the ATC system. In 2024, that stopwatch logged 16.07 million delay minutes across 359,694 reportable delays, and the cause stack is blunt: weather took 61.4% of the minutes, with volume, runway/taxiway, equipment, and the catch-all Other bucket splitting the rest. Staffing — coded only as a secondary cause under Other since mid-2022 — came to 622,669 minutes, 3.9% of the national total, yet seven facilities absorbed 86% of even that modest slice.

Those minutes are the National Airspace System's own accounting — not the passenger's gate-to-gate clock, not the airline's irregular-operations ledger — and the dashboard reads them as both a share problem and a tail problem: a decade stack and year donut under Cause mix, facility bars plus a fill-rate scatter and concentration curve under Staffing tail, and an OPSNET-vs-BTS view exposing the taxonomy trap behind "weather is only 5% of delays" headlines.

Weather still owns the majority of ATC minutes

Primary OPSNET causes are mutually exclusive at the report level: Weather, Volume, Equipment, Runway/taxiway, and Other. Weather means adverse meteorology that cuts capacity — wind, rain, snow or ice, low ceilings and visibility, thunderstorms. In 2024 weather accounted for 9.86 million minutes (61.4%). That is lower than the late-2010s weather share, when several years cleared 70%+ of minutes, but it remains the dominant primary cause every year in the 2015–2024 window except the pandemic trough.

Absolute weather minutes fell hard in 2020–2021 with traffic, then rebuilt: 7.6M (2022) → 8.7M (2023) → 9.9M (2024). The share story is slightly different from the level story. Weather's percentage of total minutes drifted down as equipment and Other grew faster off a small base, and as runway minutes jumped in 2024. Watch only the weather share and you can miss that weather minutes themselves rose more than 1.1 million year over year even while the share eased from roughly 69% (2023) to 61% (2024).

Volume, runway, and equipment are not interchangeable “non-weather”

Volume is the purity test: delays may be coded volume only when the airport is in an optimum configuration and no impacting condition is reported. In 2024 volume was 2.92 million minutes (18.2%) — the second-largest primary bucket. Volume is demand pressing against clean capacity, not a synonym for “ATC staffing.”

Runway/taxiway delays — closures and configuration changes — reached 1.65 million minutes (10.3%) in 2024, the highest runway minute total in the decade outside 2017's construction spike. Equipment remains small in share (318k minutes, 2.0%) but is no longer negligible in levels: equipment minutes roughly quadrupled from 2022 to 2024. Treating “everything that is not weather” as one residual hides that runway and equipment are infrastructure and outage stories, while volume is a schedule-vs-capacity story.

Other is small nationally — until you open the staffing secondary code

Other covers airshows, VIP restrictions, military operations, and, since mid-2022, an explicit staffing secondary cause. Other was 8.2% of 2024 minutes (1.31 million). Inside that bucket, staffing exploded: 22k minutes in 2022 → 90k in 2023 → 623k in 2024. Staffing's share of Other minutes rose from 2% to 47%. Average staffing delay lengthened from 23 to 49 minutes.

Nationally, staffing is still only ~4% of all OPSNET minutes. That is the correct headline if the question is "what share of the NAS stopwatch is a staffing code?" It is the wrong headline if the question is "where does the staffing problem live?" Concentration answers that.

Seven facilities dominate the staffing minute tail

Of 622,669 staffing minutes in 2024, 86% sat in seven facilities [National Academies OPSNET]:

• EWR226,370 minutes (AFN fill rate 87%)
• PHL183,934 minutes (fill rate 130%)
• ZNY37,953 minutes (77%)
• N9036,744 minutes (74%)
• AUS18,096 minutes (80%)
• ZJX16,860 minutes (78%)
• SFO16,035 minutes (92%)

Newark and Philadelphia alone exceed 410,000 minutes — roughly two-thirds of the national staffing total. The New York Center and New York TRACON add another ~75,000. Jacksonville Center and Austin extend the map beyond the Northeast corridor; San Francisco closes the published top seven.

PHL at 130% fill is the scatter-plot outlier. National Academies analysis ties Philadelphia's staffing-coded delays to the deliberate transfer of Newark approach control duties — a safety-driven reallocation that shows up as delay minutes even when the receiving facility is overstaffed on paper. Fill rate and delay minutes are related, not identical; the dashboard scatter is built to keep that distinction visible.

Core-30 geography already owns the delay map

Across 2022–2024, Core 30 airports plus associated TRACONs and Centers accounted for about 92% of total OPSNET delay minutes and 93% of Other minutes. Nineteen facilities staffed at 85% or less of target represented roughly 22% of operations but about 39% of total delay minutes and 45% of Other minutes. The delay map is not a uniform fog over 500 towers. It is a hub-and-center geometry with a thin staffing secondary code that is even more skewed than the primary-cause map.

ARTCC identifiers in the staffing top seven — ZNY and ZJX — matter for the “which centers dominate the tail?” question. Centers do not need to lead the national minute total to dominate the staffing tail; they appear when en-route sectors and adjacent TRACON/tower constraints couple. Towers and TRACONs (EWR, PHL, N90, AUS, SFO) still take most of the staffing minute mass in the published ranking.

OPSNET weather is not BTS “extreme weather”

Bureau of Transportation Statistics On-Time Performance uses a passenger taxonomy: air carrier, late-arriving aircraft, National Aviation System (NAS), extreme weather, and security. Extreme weather is a narrow bucket — conditions that prevent flying. Most convective weather that slows the system without grounding it is coded NAS. Late-arriving aircraft then cascades prior causes into downstream turns.

That is why passenger-facing charts can show extreme weather near ~5% of delay minutes while OPSNET shows weather near ~61% of ATC minutes in the same year [BTS On-Time Performance]. The labels collide; the definitions do not. NAS in BTS absorbs volume, non-extreme weather, airport operations, and ATC effects that OPSNET splits across Weather, Volume, Runway, Equipment, and Other. Carrier and late-aircraft categories have no OPSNET twin. Comparing the two without a taxonomy bridge produces false “weather is tiny / weather is everything” debates.

What the minute shares do — and do not — say

OPSNET minutes answer: when the ATC system detains IFR traffic for 15+ minutes, what primary cause did the facility report, and how long did it last? They do not answer passenger welfare directly, airline cost fully, or cancelation burden. They understate impacts shorter than 15 minutes. Staffing secondary coding begins mid-2022, so pre-2022 “Other” is not a clean staffing series. Average minutes rose inside Other even as counts swung — duration, not just incidence, moved.

For operations and policy readers, the actionable split is threefold. First, weather remains the majority of the NAS stopwatch; capacity and traffic-management investments that shrink weather impact still move the biggest slice. Second, runway and equipment are climbing in levels and should not be folded into a vague non-weather residual. Third, staffing is a concentrated tail: nationally small, locally decisive, with a Northeast corridor core and a handful of out-of-corridor facilities that punch above their operation share. The share question and the facility question are both true — they just describe different moments of the same OPSNET file.